mv37.org

Move 37 is a reference to AlphaGo's unconventional move in its match against Lee Sedol: a play that first looked strange, then proved decisive. We use the name because we want to help create more moments like that for complex, critical problems.

We are an independent AI research lab focused on building self-improving systems, starting with AI research itself. We are currently exploring Paperbase, which helps AI agents use frontier ideas in AI research to improve LLM training, inference, and related workflows.

We are also working on physical AI and reinforcement learning for physical systems. The goal is to build systems that can reason about the real world, learn from interaction, and improve how AI operates in environments where feedback is noisy and expensive.

A third area is automating reinforcement learning itself. We want systems that can help set up experiments, improve training loops, and make RL research faster and more reliable.

In the long term, our goal is to build systems that discover variables, infer objectives, and design experiments in pursuit of understanding. We want to construct the substrate on which AGI-like behavior can emerge: a machine that does not just learn from data, but learns how to think about the world.